Full and PEFT LoRA (LoWhIPA) fine-tuned Whisper-base and Whisper-large-v2 models for language-agnostic IPA transcription of speech.